We actually aren’t rebuilding a harness here, it’s Pi with several LLM options to select from. The reason this is a project is that the useful workflow is more like a docs test suite: run realistic user tasks across multiple models, isolate each run in a greenfield sandbox, keep the transcripts/results, and make failures reproducible in CI.

You could ask an existing coding agent to spawn subagents for every task/model pair, but once that matrix grows, running hundreds of subagents on your computer gets messy. It’s also the wrong isolation boundary: for docs testing, you usually want the agent to start from a clean environment with access only to the docs/product surface you’re testing, not your whole working tree or local setup.

So we built SkillFS. The core idea is simple: every agent sandbox is a git repo. The agent does its work, commits its progress, and the session ends. When you start a new session, SkillFS restores the repo from a git bundle and the agent continues where it left off. And because it's git, you have a complete history of every action the agent took.

The workflow looks like this:

  1. Agent works in a sandboxed environment, making changes to files
  2. Agent commits progress at meaningful checkpoints
  3. Session ends, git bundle gets saved to storage (local or GCS)
  4. Next session starts, bundle is restored, agent resumes
  5. Need to debug? git log shows you exactly what happened

  - Persistent state via git bundles, with pluggable storage backends (local filesystem or GCS), so skills and scripts from previous sessions are reusable.

  - MCP integration that lets you plug in any server. SkillFS also generates Python wrappers from MCP tool definitions and uploads them to the sandbox, so your agent can call any MCP tool as regular code to cache sequences of MCP interactions into deterministic scripts.

  - Built-in LLM runner with standard tools (glob, grep, read/write/edit files, run commands), or you can bring your own agent loop.

  - Runs on E2B sandboxes so agents execute code in isolated environments. Agent skills can also be easily imported from local files or from GitHub repos.

  > pip install skillfs

Comments URL: https://news.ycombinator.com/item?id=46543093

Points: 4

# Comments: 0